Jason

i also feel this way about glass hotel. i didnt even like station eleven all that much but glass hotel is a masterpiece.

also- this newsletter is clearly generated by chatgpt lol

i only walked 4hrs, didnt know id get something like this if i kept it up for 5 more or i would have kept going

linkedin core

what is going on

i cant believe i moved to this country man

cortado gang 😎

thinking about this again today, realized this has the funny side effect of making the Gambler’s Fallacy real

Thought I'd say a little more about some probprog-related stuff I'm building in Rust...
🧵

(If im remembering correctly, there’s a character [aptly named “spirit breaker”] with exactly this stun mechanic, the gimmick being that his draws, unlike everyone else in the game, are actually IID random)

this feels unfair under principles like “players should be able to respond to what their opponent does” and “who wins an encounter should depend more on skill and less on hitting a lucky sequence” .

maybe unfair is the wrong word! say you have an ability with a 50% chance to stun another player on hit for 1.5s. A run of (say) five or more successes could lead to a scenario where you kill the player without them ever exiting stun.

Related: a bunch of competitive games (DOTA etc) use modified non-independent processes to prevent runs making stuff unfair

observablehq.com/@manuelblanc...

I’ve been saying this since I moved here! the “space” needle is actually embarrassingly small compared to a *real* tower (like the CN tower (😎)). They use tricks of perspective in the skyline photos!

One way to solve this embarrassment would be for Seattle to build maybe 3, 4 more space needle.

same with online content spam – this stuff used to be limited by the number of people willing to write spam copy on medium or for geeksforgeeks.

lots of this stuff is transformative specifically *because* it doesn’t require large buy in.

I think most of the troubling outcomes (I think maybe most notably auto-scamming old people?) don’t require wide buy in, just a few bad actors.

not a serious post, but i do wonder who qualifies

o rly

I've played with Monte-Carlo Method... Got lost in the game

A friend of mine, I recently found out, does really fantastic memes about economics and econometrics. The initial ones are in Russian, but I convinced her to translate them into english so I could share them with a broader audience, I think many of you will appreciate them!

I actually love Zed tho, I use it for work when I need to ssh into a remote machine — it actually works better than the famously good VSCode ssh implementation ime. So impressive.

does it have repl integration or a slime equivalent yet

ime the big change happened during covid. we brought undergrads back halfway through the winter semester, and the average test grade dropped from ~85% to ~38% and never recovered. (can’t cheat in person)

this was a few months before LLMs.

haven’t wealth taxes specifically lead to capital flight in the european countries that tried them? don’t know if it’s reasonable to draw inferences from a 4% surcharge on income when wealth taxes induce effective equivalent tax rates of potentially >> 100%

my favourite gladwell incident was the Nigerian Basketball Team Hypothetical. not as obviously disqualifying as “Igon Values”, but it’s something that has just wormed its way back into my brain for years.

www.reddit.com/r/nba/commen...

thanks! yeah I’ve been mostly bussing, but it’s not efficient for longer trips (e.g. we have friends in Kenmore, and we don’t see them as much as I’d like).

For comparison, at this point, even the anemic Toronto subway system could get you as far as Kenmore from city center (I think?)

keen to read this one!

arxiv.org/abs/2504.12190
'Creating non-reversible rejection-free samplers by rebalancing skew-balanced Markov jump processes'
- Erik Jansson, Moritz Schauer, Ruben Seyer, Akash Sharma

ok so I’ve already ported most of these models to Julia (Turing). Looks like the Turing team has come out with turinglang.org/JuliaBUGS.jl..., which can plot dags

I was expecting the syntax to look like Turing essentially, making porting super easy, but it looks totally different at first glance 🤔

I think the process of creating this db was basically as you describe – they trawled the bugs manual and a bunch of textbooks for Bayesian models, and then implement them in Stan. Also includes newer stuff like Prophet, GPs, and ODE parameter inference.

GitHub - stan-dev/posteriordb: Database with posteriors of interest for Bayesian inference Database with posteriors of interest for Bayesian inference - stan-dev/posteriordb

github.com/stan-dev/pos...

if I generate dags for these with one of those automatic tools (I think they have em in numpyro(?)) would that be useful?

could probably get 80% coverage pretty easily – im sure there are some problem models.

broke: “im updating my priors”
woke: “im changing my mind”